Every object is drawn at its true size ratio. The camera distance d is stored as log₁₀(d) and changed smoothly; each frame all positions are recomputed relative to the camera in 64-bit JavaScript numbers (a floating origin) and then handed to the GPU divided by d, so the GPU only ever sees numbers between about 10⁻⁴ and 10³. Without this a 32-bit float (7 digits) could not hold 10⁻¹⁸ m and 10²⁷ m in the same scene. Near/far clip planes are fitted to the visible objects each frame.

t = size / c with c = 299 792 458 m/s (exact). For the observable universe t ≈ 93 billion years, longer than its 13.8 billion-year age because space expanded while the light travelled.

Sizes are published values (references in each card; uncertain ones are flagged). Shapes are models: the atom is a random sample of the exact hydrogen 1s density; the DNA geometry is B-form; the red blood cell uses the Evans–Fung profile; planets use procedural textures (Blender), not photographs. Schematic: spiral-arm structure, cosmic-web filaments, directions of galaxies in the Local Group, organelle placement, and the Oort cloud (hypothetical). Distances between catalogue objects are arbitrary: they are laid out on a line for the flythrough.

No experiment has found any internal structure of quarks or electrons down to ≈10⁻¹⁸ m. That bound is the starting point of this ladder. Real quarks cannot be isolated at all (confinement).

A proton, an atom, a galaxy or the Heliosphere have no sharp edge. Each size is a convention (charge radius, Bohr diameter, visible disc, termination shock…) and is stated on the card.

Models: Blender 5.2 (procedural geometry, Cycles-baked textures), Three.js r160, WebGL. Not affiliated with any organisation; no brands, no photographs.